Models to predict the mechanical properties of ternary blended fiber reinforced SCC by using mars and RVM

Abstract

This paper explores the applicability of Multivariate Adaptive Regression Splines (MARS) and Relevance Vector Machine (RVM) to predict the mechanical properties of fiber reinforced Self-Compacting Concrete (SCC) made up of quartz sand, micro silica and ground granulated blast-furnace slag (GGBS). Mechanical properties were evaluated for three grades of SCC mixes (M40, M50, and M60 with varying percentage of fibers (0.5%, 1.0%, and 1.5%) in each mix. Influencing variables were identified from the results of experimental investigations on various SCC mixes with and without fibers. Based on the output and input parameters, MARS establishes appropriate relations by using the concept of divide and conquers strategy. RVM is based on probabilistic classification utilizing the Bayesian formulation. About 70% of the mixed data of various SCC mixes was used for training after normalization of the data. The developed MARS and RVM models were verified with the rest of the data. It was found that MARS and RVM models could predict the mechanical properties of several SCC mixes very efficiently. The maximum variation between the predicted mechanical properties and the experimental observation is about ±12% for MARS model and ±15% for RVM model. Further, the efficacy of models has been tested through several statistical parameters. The developed models are useful to design the optimum experiments and for preliminary analysis of structures and components made up of similar SCC mixes.